DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to run a local agent on one sandboxed folder with Minimal Local Agent

Point a local Ollama model at a single workspace with bounded tools and edits you preview before they write.

Short answer: Clone with git clone https://github.com/MuzeAnisichael/minimal-local-agent.git and create a Python 3.11 venv; then install with python -m pip install -e . and pull a model with ollama pull qwen3.5:9b; copy the config with cp agent.example.toml agent.toml and run minimal-agent doctor. All steps.

Who it is forA developer who wants agent help on files without giving a model free rein on the disk.

The problemGeneral agents can touch anything and are hard to audit afterwards.

The outcomeAn agent confined to one workspace, with every run recorded in SQLite.

What you need first

  • Python 3.11
  • Ollama (README example pulls qwen3.5:9b) or a Chat Completions-compatible API

Steps

  1. Clone with git clone https://github.com/MuzeAnisichael/minimal-local-agent.git and create a Python 3.11 venv
  2. Install with python -m pip install -e . and pull a model with ollama pull qwen3.5:9b
  3. Copy the config with cp agent.example.toml agent.toml and run minimal-agent doctor
  4. Open the console with minimal-agent web and give it a task

Prompt for your agent

Set up MuzeAnisichael/minimal-local-agent with Ollama, run `minimal-agent doctor`, then ask it to summarize the files in <workspace>. Confirm the run appears in the local SQLite audit history and no file changed.

Caveat. Web tasks can only read or preview; approving an actual write requires the interactive CLI.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for MuzeAnisichael/minimal-local-agent and "Run a local agent on one sandboxed folder"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer